This happened between storage nodes of different clusters because they're not
informed about their state, e.g. a dead upstream storage node.

In any case, logs were flooded at 100% cpu usage.